Abstract:

In order to explore the influence mechanism of subway station passenger flow and surrounding built environment on traffic accident risk within the radiation range， the "5D+S" （5D+Subway） built environment index system was established. An accident risk model based on XG Boost algorithm and a SHAP（Shapley Additive Explanation） attribution analysis model were constructed to explore the nonlinear relationship between built environment and traffic accident risk. Taking Shenzhen as an example， this paper explores the influence mechanism of traffic accident risk around subway stations from two dimensions of weekdays and non-weekdays， and compares it with the elastic network regression model and support vector regression（SVR） model. The results show a nonlinear relationship between the built environment index of subway stations and traffic accident risk. When the density of recreational points of interest （POI） is more than 25 pieces/km²， the traffic accident risk is higher. When the accessibility of shopping malls is between [0.3，0.5]km， the traffic accident risk is higher. The built environment around subway stations has a greater impact on the risk of traffic accidents on weekdays.

Key words: around subway station, traffic accident risk, extreme gradient boosting （XG Boost） algorithm, built environment, SHAP attribution analysis